This paper presents a protocol-first specification for No-Meta intersubjective contracts: negotiation, conditional exchange, and dispute handling for autonomous agents operating under partial, local observability and without a privileged evaluator. The design is explicitly adversarial and interoperability-oriented: it remains autonomy-preserving and gaming-resistant even when counterparties are non-adopting, partially adopting, faulty, or malicious. Rather than claiming general fair exchange in an asynchronous setting without trusted third parties, the protocol targets bounded-loss, fail-closed safe withdrawal and auditable, budgeted evidence. Settlement is generalized beyond monetary payments through reservations and transfers of Value/Exposure Units (VEU) and Obligation/Task Units (OTU), plus scoped, expiring capability grants. The core artifact is an implementable contract object defined as observable-only self-binding, including terms, evidence rules, explicit self-constraints (exposure/obligation/capability caps), typed remedy rules, update rules, and an explicit Compatibility Ladder (L0–L4) that degrades safely under missing evidence, denial, or manipulation. The paper specifies deterministic packet formats (Evidence/Dispute Packets), canonicalization and digesting rules, and recommended packet chaining to resist rollback and transcript-splicing attacks. Two key correctness additions are introduced: missingness as first-class evidence (tied to channel assumptions) together with a bounded-stall liveness policy and a Progress-Safe Action Set (PSAS) for reversible actions under uncertainty; and minimal sufficient evidence as an optimization problem under an explicit information leakage cost model, enabling privacy-aware dispute packets and resisting forced over-disclosure. The specification further adds operational grounding and cost sovereignty to prevent non-operational correctness claims and coercive unbounded compute/disclosure demands.
K Takahashi (2026) studied this question.